On 23/05/07, Johnny Cheung <[EMAIL PROTECTED]> wrote:
You have mentioned that you need to reset the h/w after the block_disable(). Could you leave the MSIs enable (without calling block_disable/block_enable) while doing the h/w reset?
I tried that, but it's very difficult to avoid races between the driver interrupt handler and certain data structures which go away whilst the NIC is being reset. I really need to make sure that interrupts don't happen during this time.
Actually according to the spec, clearing the PCI_MSI_ENABLE_BIT will cause the device to trigger INTx interrupt instead of MSI. Thus, any interrupt generated after clearing the PCI_MSI_ENABLE_BIT will likely be lost. Unless you have quiesced the device and make sure no more MSI will be generated, you should not call block_disable/ddi_intr_enable.
Yes, ordinarily I would quiesce interrupts at the device first.
Also from your debug output, I noticed that the device has two MSIs and their corresponding vectors are 0x62 and 0x63 after the first initial block enable. While the 2nd time, the MSIs have vectors 0x61 and 0x62 which are not aligned properly. There is a bug using the non-aligned based vector for multi-MSIs but it was fixed in snv_44. Since you are using a much newer build, you shouldn't encounter this problem. Anyway, I will have to look into the codes and get back to you on this. In the mean time, just to make sure, instead of using block_enable/block_disable, could you change that to ddi_intr_enable/ddi_intr_disable with a for loop? It should do the same job for you.
I can't use ddi_intr_disable/enable because the 'block' capability is set, so these functions return DDI_EINVAL. It has to be a block disable/enable. I can work around the problem by tearing things down and re-allocating the interrupt handles each time so it's not a showstopper for me; but it does seem like a bug. Paul -- Paul Durrant http://www.linkedin.com/in/pdurrant _______________________________________________ driver-discuss mailing list driver-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/driver-discuss
